Subgrouping Germinal Center-Derived B-Cell Lymphomas based on Machine Learning-deduced DNA Methylation Modules

Despite all being derived from germinal center B-cells, follicular and diffuse large B-cell lymphomas are biologically and clinically heterogeneous. DNA methylation has emerged as a robust biomarker for classification of solid tumors, but its application to subtype mature B-cell lymphomas is still in its beginning. Therefore, we conducted array-based DNA methylation analyses on 177 molecularly well-characterized mature B-cell lymphomas from the ICGC MMML-Seq project. The Phenotype-Genotype Many-to-Many Relations Analysis machine learning method was applied to identify significant biclusters of CpGs and samples. The analysis revealed 300 CpGs forming four modules, which ordered the lymphomas into seven methylation patterns (MP1-7). These MP1-7 showed significant associations with biological features of the lymphomas. For example, MP1 and MP2 both predominately comprised follicular lymphomas but significantly differed in age at diagnosis and proliferation history. MP4-MP7 consisted mainly of diffuse large B-cell lymphomas but showed differences regarding their cell-of-origin signatures. MP6 showed enrichment of cases with an MCD/C5-like mutational signature with MYD88 alterations as hallmark. The 300 CpGs also segregated apart Burkitt lymphoma and non-malignant B-cell populations. The identified CpG modules and methylation profiles, thus, reflect properties of mature B-cell lymphomas across different biologic layers from age at diagnosis to mutational signatures. We collected 177 samples (cryo-preserved) from patients with follicular lymphoma (FL), diffuse large B cell lymphoma (DLBCL) and follicular lymphoma with diffuse large B cell lymphoma (FL-DLBCL) (cryo-preserved material) from the ICGC MMML-Seq consortium. The samples were analyzed using the Infinium HumanMethylation450 BeadChip array.